Traditional camera sensors rely on human eyes for observation. However, human eyes are prone to fatigue when observing objects of different sizes for a long time in complex scenes, and human cognition is limited, which often leads to judgment errors and greatly reduces efficiency. Object recognition technology is an important technology used to judge the object’s category on a camera sensor. In order to solve this problem, a small-size object detection algorithm for special scenarios was proposed in this paper. The advantage of this algorithm is that it not only has higher precision for small-size object detection but also can ensure that the detection accuracy for each size is not lower than that of the existing algorithm. There are three main innovations in this paper, as follows: (1) A new downsampling method which could better preserve the context feature information is proposed. (2) The feature fusion network is improved to effectively combine shallow information and deep information. (3) A new network structure is proposed to effectively improve the detection accuracy of the model. From the point of view of detection accuracy, it is better than YOLOX, YOLOR, YOLOv3, scaled YOLOv5, YOLOv7-Tiny, and YOLOv8. Three authoritative public datasets are used in these experiments: (a) In the Visdron dataset (small-size objects), the map, precision, and recall ratios of DC-YOLOv8 are 2.5%, 1.9%, and 2.1% higher than those of YOLOv8s, respectively. (b) On the Tinyperson dataset (minimal-size objects), the map, precision, and recall ratios of DC-YOLOv8 are 1%, 0.2%, and 1.2% higher than those of YOLOv8s, respectively. (c) On the PASCAL VOC2007 dataset (normal-size objects), the map, precision, and recall ratios of DC-YOLOv8 are 0.5%, 0.3%, and 0.4% higher than those of YOLOv8s, respectively.

As the road traffic situation becomes complex, the task of traffic management takes on an increasingly heavy load. The air-to-ground traffic administration network of drones has become an important tool to promote the high quality of traffic police work in many places. Drones can be used instead of a large number of human beings to perform daily tasks, as: traffic offense detection, daily crowd detection, etc. Drones are aerial operations and shoot small targets. So the detection accuracy of drones is less. To address the problem of low accuracy of Unmanned Aerial Vehicles (UAVs) in detecting small targets, we designed a more suitable algorithm for UAV detection and called GBS-YOLOv5. It was an improvement on the original YOLOv5 model. Firstly, in the default model, there was a problem of serious loss of small target information and insufficient utilization of shallow feature information as the depth of the feature extraction network deepened. We designed the efficient spatio-temporal interaction module to replace the residual network structure in the original network. The role of this module was to increase the network depth for feature extraction. Then, we added the spatial pyramid convolution module on top of YOLOv5. Its function was to mine small target information and act as a detection head for small size targets. Finally, to better preserve the detailed information of small targets in the shallow features, we proposed the shallow bottleneck. And the introduction of recursive gated convolution in the feature fusion section enabled better interaction of higher-order spatial semantic information. The GBS-YOLOv5 algorithm conducted experiments showing that the value of mAP@0.5 was 35.3$$\%$$ % and the mAP@0.5:0.95 was 20.0$$\%$$ % . Compared to the default YOLOv5 algorithm was boosted by 4.0$$\%$$ % and 3.5$$\%$$ % , respectively.

Driven by deep learning, great breakthroughs had been made in the field of target detection. Small target detection algorithms were widely used in industry, agriculture and other fields. But the small target had few available features and the loss of small target detail information in feature extraction. So it led to the low accuracy of the small target detection algorithms. In this paper, we proposed DBF‐YOLO algorithm based on the classical YOLOV5. The classical YOLOV5 algorithm with high speed. The detection speed of the minimum model could reach 24 ms. However, the deep network structure led to the low detection accuracy of small targets. Our proposed DBF‐YOLO algorithm was an improvement on the problem of small target information being lost. The main contributions of this article were mainly: First, a shallow feature extraction network was introduced in P1 layer, more details of small targets could be well retained. Second, by adding the feature fusion network of shallow feature map and the detection output part in the FPN + PAN layers, the algorithm's accuracy and generalization ability were significantly enhanced. Compared to YOLOV5, the performance of the DBF‐YOLO algorithm was significantly improved. On the validation set, mAP@0.5 and mAP@0.5:0.95 were increased by 8.80 and 5.90%, respectively. Recall was increased from the initial 34.50–41.80%. Precision was increased from initial 44.20 to 50.70%. On the test set, mAP@0.5 and mAP@0.5:0.95 were increased by 6.40 and 3.90%, respectively. Recall was increased 5.10%. Precision was increased 6.60%. Experiments had shown that the improved algorithm achieved good results in accuracy. © 2023 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.